- KonferenzbeitragShallow CNNs for the Reliable Detection of Facial Marks(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Zeinstra, Chris; Haasnoot, Erwin; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasFacial marks are local irregularities of skin texture. Their type and/or spatial pattern can be used as a (soft) biometric modality in several applications. A key requirement for a biometric system that utilises facial marks is their reliable detection. Detection methods typically use a blob detector followed by heuristic post processing steps to reduce the number of false positives. In this paper, we consider shallow Convolutional Neural Networks (CNNs) for facial mark detection. The choice of this network type seems natural as it learns multiple (non) blob detectors; shallow refers to the fact that we only consider CNNs up to three layers.We show that (a) these CNNs successfully address the false positive problem, (b) remove the need for post processing steps, and (c) outperform a classic blob detector, approaches taken in previous studies and some other non CNN type classifiers in terms of EER and FMR at TMR=0.95.
- KonferenzbeitragDeep Domain Adaptation for Face Recognition using images captured from surveillance cameras(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Banerjee, Samik; Bhattacharjee, Avishek; Das, Sukhendu; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasLearning based on convolutional neural networks (CNNs) or deep learning has been a major research area with applications in face recognition (FR). However, performances of algorithms designed for FR are unsatisfactory when surveillance conditions severely degrade the test probes. The work presented in this paper has three contributions. First, it proposes a novel adaptive-CNN architecture of deep learning refurbished for domain adaptation (DA), to overcome the difference in feature distributions between the gallery and probe samples. The proposed architecture consists of three components: feature (FM), adaptive (AM) and classification (CM) modules. Secondly, a novel 2-stage algorithm for Mutually Exclusive Training (2-MET) based on stochastic gradient descent, has been proposed. The final stage of training in 2-MET freezes the layers of the FM and CM, while updating (tuning) only the parameters of the AM using a few probe (as target) samples. This helps the proposed deep-DA CNN to bridge the disparities in the distributions of the gallery and probe samples, resulting in enhanced domain-invariant representation for efficient deep-DA learning and classification. The third contribution comes from rigorous experimentations performed on three benchmark real-world surveillance face datasets with various kinds of degradations. This reveals the superior performance of the proposed adaptive-CNN architecture with 2-MET training, using Rank-1 recognition rates and ROC and CMC metrics, over many recent state-of-the-art techniques of CNN and DA.
- KonferenzbeitragUnsupervised Facial Geometry Learning for Sketch to Photo Synthesis(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Kazemi, Hadi; Taherkhani, Fariborz; Nasrabadi, Nasser M.; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasFace sketch-photo synthesis is a critical application in law enforcement and digital entertainment industry where the goal is to learn the mapping between a face sketch image and its corresponding photo-realistic image. However, the limited number of paired sketch-photo training data usually prevents the current frameworks to learn a robust mapping between the geometry of sketches and their matching photo-realistic images. Consequently, in this work, we present an approach for learning to synthesize a photo-realistic image from a face sketch in an unsupervised fashion. In contrast to current unsupervised image-to-image translation techniques, our framework leverages a novel perceptual discriminator to learn the geometry of human face. Learning facial prior information empowers the network to remove the geometrical artifacts in the face sketch.We demonstrate that a simultaneous optimization of the face photo generator network, employing the proposed perceptual discriminator in combination with a texture-wise discriminator, results in a significant improvement in quality and recognition rate of the synthesized photos. We evaluate the proposed network by conducting extensive experiments on multiple baseline sketch-photo datasets.
- KonferenzbeitragFake Face Detection Methods: Can They Be Generalized?(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Khodabakhsh, Ali; Ramachandra, Raghavendra; Raja, Kiran; Wasnik, Pankaj; Busch, Christoph; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasWith advancements in technology, it is now possible to create representations of human faces in a seamless manner for fake media, leveraging the large-scale availability of videos. These fake faces can be used to conduct personation attacks on the targeted subjects. Availability of open source software and a variety of commercial applications provides an opportunity to generate fake videos of a particular target subject in a number of ways. In this article, we evaluate the generalizability of the fake face detection methods through a series of studies to benchmark the detection accuracy. To this extent, we have collected a new database of more than 53;000 images, from 150 videos, originating from multiple sources of digitally generated fakes including Computer Graphics Image (CGI) generation and many tampering based approaches. In addition, we have also included images (with more than 3;200) from the predominantly used Swap-Face application that is commonly available on smart-phones. Extensive experiments are carried out using both texture-based handcrafted detection methods and deep learning based detection methods to find the suitability of detection methods. Through the set of evaluation, we attempt to answer if the current fake face detection methods can be generalizable.
- KonferenzbeitragFingerprint Presentation Attack Detection using Laser Speckle Contrast Imaging(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Keilbach, Pascal; Kolberg, Jascha; Gomez-Barrero, Marta; Busch, Christoph; Langweg, Hanno; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasWith the increased deployment of biometric authentication systems, some security concerns have also arisen. In particular, presentation attacks directed to the capture device pose a severe threat. In order to prevent them, liveness features such as the blood flow can be utilised to develop presentation attack detection (PAD) mechanisms. In this context, laser speckle contrast imaging (LSCI) is a technology widely used in biomedical applications in order to visualise blood flow. We therefore propose a fingerprint PAD method based on textural information extracted from preprocessed LSCI images. Subsequently, a support vector machine is used for classification. In the experiments conducted on a database comprising 32 different artefacts, the results show that the proposed approach classifies correctly all bona fides. However, the LSCI technology experiences difficulties with thin and transparent overlay attacks.
- KonferenzbeitragDeep Sparse Feature Selection and Fusion for Textured Contact Lens Detection(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Poster, Domenick; Nasrabadi, Nasser; Riggan, Benjamin; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasDistinguishing between images of irises wearing textured lenses versus those wearing transparent lenses or no lenses is a challenging problem due to the subtle and fine-grained visual differences. Our approach builds upon existing hand-crafted image features and neural network architectures by optimally selecting and combining the most useful set of features into a single model. We build multiple, parallel sub-networks corresponding to the various feature descriptors and learn the best subset of features through group sparsity. We avoid overfitting such a wide and deep model through a selective transfer learning technique and a novel group Dropout regularization strategy. This model achieves roughly a four times increase in performance over the state-of-the-art on three benchmark textured lens datasets and equals the near-perfect state-of-the-art accuracy on two others. Furthermore, the generic nature of the architecture allows it to be extended to other image features, forms of spoofing attacks, or problem domains.
- KonferenzbeitragMobiBits: Multimodal Mobile Biometric Database(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Bartuzi, Ewelina; Roszczewska, Katarzyna; Trokielewicz, Mateusz; Białobrzeski, Radosław; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasThis paper presents a novel database comprising representations of five different biometric characteristics, collected in a mobile, unconstrained or semi-constrained setting with three different mobile devices, including characteristics previously unavailable in existing datasets, namely hand images, thermal hand images, and thermal face images, all acquired with a mobile, off-the-shelf device. In addition to this collection of data we perform an extensive set of experiments providing insight on benchmark recognition performance that can be achieved with these data, carried out with existing commercial and academic biometric solutions. This is the first known to us mobile biometric database introducing samples of biometric traits such as thermal hand images and thermal face images. We hope that this contribution will make a valuable addition to the already existing databases and enable new experiments and studies in the field of mobile authentication. The MobiBits database is made publicly available to the research community at no cost for non-commercial purposes.
- KonferenzbeitragFEERCI: A Package for Fast Non-Parametric Confidence Intervals for Equal Error Rates in Amortized O(m log n)(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Haasnoot, Erwin; Khodabakhsh, Ali; Zeinstra, Chris; Spreeuwers, Luuk; Veldhuis, Raymond; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasEqual Error Rates (EERs), or other weighted relations between False Match and Non- Match Rates (FMR/FNMR), are often used as a performance metric for biometric systems. Confidence Intervals (CIs) are used to denote the uncertainty underlying these EERs, with many methods existing to estimate said CIs in both parametric and non-parametric ways. These confidence intervals provide, foremost, a method of comparing scoring/ranking functions. Non-parametric methods often suffer from high computational costs, but do not make assumptions as to the shape of the EERand score distributions. For both EERs and CIs, contemporary open-source toolkits leave room for improvement in terms of computational efficiency. In this paper, we introduce the Fast EER (FEER) algorithm that calculates an EER in O(logn) on a sorted score list, we show how to adapt the FEER algorithm to calculate non-parametric, bootstrapped EER CIs (FEERCI) in O(mlogn) given m resamplings, and we introduce an opinionated open-source package named feerci that provides implementations of the FEER and FEERCI algorithm.We provide speed and accuracy benchmarks for the feerci package, comparing it against the most-used methods of calculating EERs in Python and show how it is able to calculate EERs and CIs on very large score lists faster than contemporary toolkits can calculate a single EER.
- KonferenzbeitragGait template protection using HMM-UBM(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Van hamme, Tim; Argones Rúa, Enrique; Preuveneers, Davy; Joosen, Wouter; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasThis paper presents a hidden Markov model-Universal background model gait authentication system, which is also incorporated into a template protection based on a fuzzy commitment scheme.We show that with limited enrollment data the HMM-UBM system achieves a very competitive equal error rate of 1% using one sensor. The proposed template protection scheme benefits from eigenfeatures coming from multiple Universal background model systems fused with a novel technique that minimizes the bit error rate for genuine attempts. This allows the protected system to achieve a false rejection rate below 5% with an effective key length of 64 bits.
- KonferenzbeitragUnsupervised Learning of Fingerprint Rotations(BIOSIG 2018 - Proceedings of the 17th International Conference of the Biometrics Special Interest Group, 2018) Schuch, Patrick; May, Jan Marek; Busch, Christoph; Brömme, Arslan; Busch, Christoph; Dantcheva, Antitza; Rathgeb, Christian; Uhl, AndreasThe alignment of fingerprint samples is a preprocessing step in fingerprint recognition. It allows an improved biometric feature extraction and a more accurate biometric comparison. We propose to use Convolutional Neural Networks for estimation of the rotational part. The main contribution is an unsupervised training strategy similar to Siamese Networks for estimation of rotations. The approach does not need any labelled data for training. It is trained to estimate orientation differences for pairs of samples. Our approach achieves an alignment accuracy with a mean absolute deviation 2:1 on data similar to the training data, which supports the alignment task. For other datasets accuracies down to 6:2 mean absolute deviation are achieved.